Naslov
Building a labeled dataset for recognition of handball actions using mask R-CNN and STIPS

Autori
Ivašić-Kos, Marina ; Pobar, Miran

Vrsta, podvrsta i kategorija rada
Radovi u zbornicima skupova, cjeloviti rad (in extenso), znanstveni

Izvornik
2018 7th European Workshop on Visual Information Processing (EUVIP) / - : Institute of Electrical and Electronics Engineers (IEEE), 2018, 1-6

ISBN
978-1-5386-6897-9

Skup
7th European Workshop on Visual Information Processing (EUVIP 2018)

Mjesto i datum
Tampere, Finska, 26.11.2018. - 28.11.2018

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
object detectors , sports scenes , Mask R-CNN , spatio-temporal interest point-STIP , action recognition database

Sažetak
Building successful machine learning models depends on large amounts of training data that often needs to be labelled manually. We propose a method to efficiently build an action recognition dataset in the handball domain, focusing on minimizing the manual labor required to label the individual players performing the chosen actions. The method uses existing deep learning object recognition methods for player detection and combines the obtained location information with a player activity measure based on spatio-temporal interest points to track players that are performing the currently relevant action, here called active players. The method was successfully used on a challenging dataset of real-world handball practice videos, where the leading active player was correctly tracked and labeled in 84 % of cases.

Izvorni jezik
Engleski

Znanstvena područja
Računarstvo, Informacijske i komunikacijske znanosti

POVEZANOST RADA